host/taint: uncertainty is not history — it never enters sticky state
Found by the phase-5 audit on the live graph, immediately after phase 3r landed: a hardware sink carried a permanent `unresolved-ancestry` taint. The cause was one link observed while its output node was still unbound — a correct fail-closed answer — which was then written into sticky state, where retirement requires every member object to be absent. A live sound card never is, so the mark survived readiness, 21 recomputes and deliberate churn. Phase 3r makes this systematic rather than rare: every node is now withheld until its bind resolves, so any link seen across that gap raises `UnresolvedAncestry` on its input side. It fires at startup, every startup. User decision (2026-07-25): uncertainty-based taint retires once the uncertainty is gone; evidence-based taint keeps the absence rule. Retiring by reason *code* would not be enough, because uncertainty launders itself — an unresolved node propagates `TaintedUpstream`, which is indistinguishable from real contamination once recorded. So the split is by **provenance**: `evaluate` runs the fixpoint twice. Pass 1 fails closed exactly as before and is what every decision is made from; pass 2 raises no uncertainty root at all, and is the only thing sticky state is built from. Nothing derived from an uncertainty can reach the sticky path. Decisions are unchanged by construction — all 57 existing taint tests pass untouched, including the fail-closed and sticky-survival rows. 3 new tests, mutation-verified (pointing `build_sticky` back at the fail-closed taint kills exactly the two new uncertainty tests and nothing else): unresolved ancestry clears once resolved; taint laundered downstream of an uncertainty clears with it; real taint still survives its topology disappearing. Live: the audit's post-readiness records now report taint 0 where they reported a permanent sticky entry before. Recompute cost roughly doubles as expected (two fixpoints) — 80 µs worst case observed, against a 47 Hz event rate. `taint/tests.rs` keeps its one pre-existing hand-formatted line; everything else in both files is rustfmt-clean.
This commit is contained in:
+129
-34
@@ -381,36 +381,28 @@ pub fn evaluate(
|
||||
let components = OwnerComponents::build(snapshot, ctx.pipewire_pulse_pid);
|
||||
let keys = owner::OwnerKeyIndex::build(snapshot, ctx.pipewire_pulse_pid);
|
||||
|
||||
let mut taint: BTreeMap<Serial, Reason> = BTreeMap::new();
|
||||
let mut sticky_serials: BTreeSet<Serial> = BTreeSet::new();
|
||||
|
||||
seed_local_roots(snapshot, ctx, &mut taint);
|
||||
seed_sticky(
|
||||
// Pass 1 — the fail-closed view. Every decision is made from this one, so
|
||||
// "we could not see" counts as taint.
|
||||
let (taint, sticky_serials) = compute_taint(
|
||||
snapshot,
|
||||
ctx,
|
||||
&keys,
|
||||
prior,
|
||||
&components,
|
||||
&mut taint,
|
||||
&mut sticky_serials,
|
||||
prior,
|
||||
Uncertainty::FailsClosed,
|
||||
);
|
||||
|
||||
// Monotone fixpoint: every step only adds taint, or lowers a node's
|
||||
// reason priority, both of which are bounded. Link propagation and the
|
||||
// owner bridge feed each other — a bridged output leg has downstream
|
||||
// links, and a downstream monitor reader bridges to its own siblings —
|
||||
// so neither can be run once.
|
||||
let edges = downstream_edges(snapshot, &mut taint);
|
||||
loop {
|
||||
let mut changed = false;
|
||||
changed |= propagate_links(&edges.edges, &mut taint);
|
||||
changed |= propagate_owner_bridge(&keys, &components, &edges, &mut taint);
|
||||
changed |= propagate_unresolved_owner(snapshot, &keys, &edges, &mut taint);
|
||||
if !changed {
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
let decisions = build_decisions(snapshot, ctx, &taint, &sticky_serials);
|
||||
|
||||
// Pass 2 — the evidence-only view, and the only thing sticky state is
|
||||
// ever built from (see [`Uncertainty`]).
|
||||
let (evidence, _) = compute_taint(
|
||||
snapshot,
|
||||
ctx,
|
||||
&keys,
|
||||
&components,
|
||||
prior,
|
||||
Uncertainty::Ignored,
|
||||
);
|
||||
// ⚠️ Readiness gates **retirement only**, never addition (Codex rounds
|
||||
// 1 and 2, which caught the two halves of this in turn). An object
|
||||
// missing from an untrustworthy snapshot has not been observed to
|
||||
@@ -419,10 +411,105 @@ pub fn evaluate(
|
||||
// *observed* during a not-ready epoch is real — a reader can consume
|
||||
// and buffer the call and then vanish before readiness — so discarding
|
||||
// additions was the same defect pointing the other way.
|
||||
let next_sticky = build_sticky(snapshot, &keys, &components, &taint, prior, ctx.graph_ready);
|
||||
let next_sticky = build_sticky(
|
||||
snapshot,
|
||||
&keys,
|
||||
&components,
|
||||
&evidence,
|
||||
prior,
|
||||
ctx.graph_ready,
|
||||
);
|
||||
(decisions, next_sticky)
|
||||
}
|
||||
|
||||
/// Whether a pass treats "we could not see" as taint.
|
||||
///
|
||||
/// **Both passes exist because stickiness is a claim about history, and
|
||||
/// uncertainty is not history.** A node tainted only because the graph was
|
||||
/// mid-enumeration has had nothing observed about it; remembering that as
|
||||
/// taint forever is over-exclusion with no evidence behind it, and phase 3r's
|
||||
/// bind-everything observer makes the window it happens in systematically
|
||||
/// wide (every node is withheld until its bind resolves, so any link observed
|
||||
/// across that gap raises [`Reason::UnresolvedAncestry`] on its input side).
|
||||
/// Measured on a live desktop: a hardware sink acquired a permanent sticky
|
||||
/// taint at every startup, from one link seen while its output node was still
|
||||
/// unbound.
|
||||
///
|
||||
/// Retiring by *reason code* is not enough, because uncertainty launders
|
||||
/// itself: an unresolved node propagates [`Reason::TaintedUpstream`] to its
|
||||
/// downstream, and that reason is indistinguishable from real contamination
|
||||
/// once recorded. So the split is by **provenance** — the sticky pass never
|
||||
/// raises an uncertainty root at all, and nothing derived from one can reach
|
||||
/// it. Decisions are unaffected: they are made from the fail-closed pass,
|
||||
/// which is unchanged.
|
||||
#[derive(Clone, Copy, PartialEq, Eq, Debug)]
|
||||
enum Uncertainty {
|
||||
/// Unresolved ancestry and an unbounded tainted reader are taint
|
||||
/// (v3.4 §6.1, §6.1.1, §6.1.4).
|
||||
FailsClosed,
|
||||
/// Only positively observed contamination counts.
|
||||
Ignored,
|
||||
}
|
||||
|
||||
/// One taint fixpoint over the snapshot. The `uncertainty` mode decides
|
||||
/// whether absence of evidence is treated as evidence of contamination.
|
||||
fn compute_taint(
|
||||
snapshot: &GraphSnapshot,
|
||||
ctx: &ExclusionCtx,
|
||||
keys: &owner::OwnerKeyIndex,
|
||||
components: &OwnerComponents,
|
||||
prior: &StickyState,
|
||||
uncertainty: Uncertainty,
|
||||
) -> (BTreeMap<Serial, Reason>, BTreeSet<Serial>) {
|
||||
let fails_closed = uncertainty == Uncertainty::FailsClosed;
|
||||
let mut taint: BTreeMap<Serial, Reason> = BTreeMap::new();
|
||||
let mut sticky_serials: BTreeSet<Serial> = BTreeSet::new();
|
||||
|
||||
seed_local_roots(snapshot, ctx, &mut taint);
|
||||
if fails_closed {
|
||||
for serial in ambiguous_id_nodes(snapshot) {
|
||||
raise(&mut taint, serial, Reason::UnresolvedAncestry);
|
||||
}
|
||||
}
|
||||
seed_sticky(
|
||||
snapshot,
|
||||
keys,
|
||||
prior,
|
||||
components,
|
||||
&mut taint,
|
||||
&mut sticky_serials,
|
||||
);
|
||||
|
||||
// Edges are built identically in both passes — receiver status is a
|
||||
// topological fact and must not depend on the mode, or the owner bridge
|
||||
// would see two different graphs.
|
||||
let mut unresolved_input: BTreeSet<Serial> = BTreeSet::new();
|
||||
let edges = downstream_edges(snapshot, &mut unresolved_input);
|
||||
if fails_closed {
|
||||
for serial in unresolved_input {
|
||||
raise(&mut taint, serial, Reason::UnresolvedAncestry);
|
||||
}
|
||||
}
|
||||
|
||||
// Monotone fixpoint: every step only adds taint, or lowers a node's
|
||||
// reason priority, both of which are bounded. Link propagation and the
|
||||
// owner bridge feed each other — a bridged output leg has downstream
|
||||
// links, and a downstream monitor reader bridges to its own siblings —
|
||||
// so neither can be run once.
|
||||
loop {
|
||||
let mut changed = false;
|
||||
changed |= propagate_links(&edges.edges, &mut taint);
|
||||
changed |= propagate_owner_bridge(keys, components, &edges, &mut taint);
|
||||
if fails_closed {
|
||||
changed |= propagate_unresolved_owner(snapshot, keys, &edges, &mut taint);
|
||||
}
|
||||
if !changed {
|
||||
break;
|
||||
}
|
||||
}
|
||||
(taint, sticky_serials)
|
||||
}
|
||||
|
||||
/// Roots that are visible on the node itself.
|
||||
fn seed_local_roots(
|
||||
snapshot: &GraphSnapshot,
|
||||
@@ -433,14 +520,20 @@ fn seed_local_roots(
|
||||
if let Some(reason) = local_root_reason(node, ctx) {
|
||||
raise(taint, node.serial, reason);
|
||||
}
|
||||
// A node whose own global id is ambiguous cannot be the reliable
|
||||
// endpoint of any link, so its ancestry is unresolvable.
|
||||
if snapshot.node_by_id(node.id) == Some(IdLookup::Ambiguous) {
|
||||
raise(taint, node.serial, Reason::UnresolvedAncestry);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Nodes whose own global id is ambiguous: they cannot be the reliable
|
||||
/// endpoint of any link, so their ancestry is unresolvable. Uncertainty, not
|
||||
/// evidence — see [`Uncertainty`].
|
||||
fn ambiguous_id_nodes(snapshot: &GraphSnapshot) -> BTreeSet<Serial> {
|
||||
snapshot
|
||||
.nodes()
|
||||
.filter(|node| snapshot.node_by_id(node.id) == Some(IdLookup::Ambiguous))
|
||||
.map(|node| node.serial)
|
||||
.collect()
|
||||
}
|
||||
|
||||
fn local_root_reason(node: &NodeSnapshot, ctx: &ExclusionCtx) -> Option<Reason> {
|
||||
if node.props.peerspeak_owned {
|
||||
return Some(Reason::PeerspeakOwned);
|
||||
@@ -563,7 +656,7 @@ fn nodes_of_client(
|
||||
/// `output node → input nodes`, resolving snapshot-local ids. An endpoint
|
||||
/// that does not resolve taints the *other* end as unresolved ancestry when
|
||||
/// that other end is the input side — we cannot know what is feeding it.
|
||||
fn downstream_edges(snapshot: &GraphSnapshot, taint: &mut BTreeMap<Serial, Reason>) -> Edges {
|
||||
fn downstream_edges(snapshot: &GraphSnapshot, unresolved_input: &mut BTreeSet<Serial>) -> Edges {
|
||||
let mut edges: BTreeMap<Serial, Vec<Serial>> = BTreeMap::new();
|
||||
let mut receivers: BTreeSet<Serial> = BTreeSet::new();
|
||||
for link in snapshot.links() {
|
||||
@@ -575,8 +668,10 @@ fn downstream_edges(snapshot: &GraphSnapshot, taint: &mut BTreeMap<Serial, Reaso
|
||||
receivers.insert(to);
|
||||
}
|
||||
(_, Some(IdLookup::Unique(to))) => {
|
||||
// Something feeds this node and we cannot say what.
|
||||
raise(taint, to, Reason::UnresolvedAncestry);
|
||||
// Something feeds this node and we cannot say what. Reported
|
||||
// rather than raised here, because whether "cannot say" is
|
||||
// taint depends on which pass is running ([`Uncertainty`]).
|
||||
unresolved_input.insert(to);
|
||||
receivers.insert(to);
|
||||
}
|
||||
(_, Some(IdLookup::Ambiguous)) => {
|
||||
|
||||
@@ -820,6 +820,137 @@ fn an_ambiguous_recycled_global_id_fails_closed() {
|
||||
);
|
||||
}
|
||||
|
||||
// ──────────────────────────────────────────────────────────────────────
|
||||
// Uncertainty is not history — it never enters sticky state
|
||||
// (round 9, from a live phase-5 audit run; see `Uncertainty` in mod.rs)
|
||||
// ──────────────────────────────────────────────────────────────────────
|
||||
|
||||
#[test]
|
||||
fn unresolved_ancestry_does_not_survive_being_resolved() {
|
||||
// Measured live on a desktop: a link is observed while its output node is
|
||||
// still unbound, the input side fails closed — correctly — and then that
|
||||
// fail-closed mark became *sticky*, so a hardware sink stayed excluded for
|
||||
// the process lifetime even after the node resolved and turned out to be
|
||||
// an ordinary game. Phase 3r's bind-everything observer widens that window
|
||||
// to every node, so this must clear.
|
||||
let mut graph = Graph::new();
|
||||
let ghost = graph.dangling_id();
|
||||
let client = graph.client_of_app(6000);
|
||||
let victim = graph.node("victim-in", MediaRole::StreamInput, app(client, 6000));
|
||||
let sibling = graph.node("victim-out", MediaRole::StreamOutput, app(client, 6000));
|
||||
graph.link_ids(ghost, victim.id);
|
||||
let firefox = graph.app_node("firefox", MediaRole::StreamOutput, 11114);
|
||||
let c = ctx();
|
||||
|
||||
// While the ancestry is genuinely unresolved, the decision is unchanged:
|
||||
// fail closed, both the victim and its sibling excluded.
|
||||
let (first, sticky) = evaluate(&graph.build(), &c, &StickyState::default());
|
||||
assert_partition(
|
||||
&first,
|
||||
&[("firefox", firefox)],
|
||||
&[("victim-out", sibling, "tainted-owner-bridge")],
|
||||
);
|
||||
assert_tainted(&first, victim, "unresolved-ancestry");
|
||||
|
||||
// The node behind that id turns up — nothing tainted, it was simply not
|
||||
// observed yet. The uncertainty is gone, so nothing may remain of it.
|
||||
let late_client = graph.client_of_app(7100);
|
||||
let resolved = graph.node_with_id(
|
||||
"was-unbound",
|
||||
MediaRole::StreamOutput,
|
||||
ghost,
|
||||
app(late_client, 7100),
|
||||
);
|
||||
let (second, _) = evaluate(&graph.build(), &c, &sticky);
|
||||
assert_partition(
|
||||
&second,
|
||||
&[
|
||||
("firefox", firefox),
|
||||
("victim-out", sibling),
|
||||
("was-unbound", resolved),
|
||||
],
|
||||
&[],
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn uncertainty_laundered_into_downstream_taint_is_not_sticky_either() {
|
||||
// Retiring by reason *code* would not be enough: an unresolved node
|
||||
// propagates `tainted-upstream`, which is indistinguishable from real
|
||||
// contamination once recorded. The split has to be by provenance, so a
|
||||
// node two hops from the uncertainty must clear too.
|
||||
let mut graph = Graph::new();
|
||||
let ghost = graph.dangling_id();
|
||||
let forwarder_client = graph.client_of_app(6100);
|
||||
let forwarder_in = graph.node(
|
||||
"fwd-in",
|
||||
MediaRole::StreamInput,
|
||||
app(forwarder_client, 6100),
|
||||
);
|
||||
let forwarder_out = graph.node(
|
||||
"fwd-out",
|
||||
MediaRole::StreamOutput,
|
||||
app(forwarder_client, 6100),
|
||||
);
|
||||
let downstream_client = graph.client_of_app(6200);
|
||||
let downstream = graph.node("downstream", MediaRole::Sink, app(downstream_client, 6200));
|
||||
let downstream_leg = graph.node(
|
||||
"downstream-out",
|
||||
MediaRole::StreamOutput,
|
||||
app(downstream_client, 6200),
|
||||
);
|
||||
graph.link_ids(ghost, forwarder_in.id);
|
||||
graph.link(forwarder_out, downstream);
|
||||
let c = ctx();
|
||||
|
||||
let (first, sticky) = evaluate(&graph.build(), &c, &StickyState::default());
|
||||
assert_tainted(&first, forwarder_in, "unresolved-ancestry");
|
||||
assert_tainted(&first, downstream, "tainted-upstream");
|
||||
assert!(
|
||||
first.candidates[&downstream_leg.serial].reason().is_some(),
|
||||
"while the ancestry is unresolved the downstream owner is excluded too"
|
||||
);
|
||||
|
||||
let late_client = graph.client_of_app(7200);
|
||||
graph.node_with_id(
|
||||
"was-unbound",
|
||||
MediaRole::StreamOutput,
|
||||
ghost,
|
||||
app(late_client, 7200),
|
||||
);
|
||||
let (second, _) = evaluate(&graph.build(), &c, &sticky);
|
||||
assert_eq!(
|
||||
second.candidates[&downstream_leg.serial].reason(),
|
||||
None,
|
||||
"nothing derived from the uncertainty may outlive it"
|
||||
);
|
||||
assert_eq!(
|
||||
second.candidates[&forwarder_out.serial].reason(),
|
||||
None,
|
||||
"including the unresolved node's own owner siblings"
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn real_taint_is_still_sticky_when_its_topology_goes_away() {
|
||||
// The other half of the same rule, stated positively: *evidence* is
|
||||
// history and must survive. This is the guard on the change above — if
|
||||
// provenance splitting ever leaks into the evidence path, peerspeak's own
|
||||
// audio starts escaping.
|
||||
let (graph, call, rec_in, rec_out, firefox) = sticky_scene();
|
||||
let c = ctx();
|
||||
let (_, sticky) = evaluate(&graph.build(), &c, &StickyState::default());
|
||||
let (second, _) = evaluate(&graph.build_without(&[rec_in]), &c, &sticky);
|
||||
assert_partition(
|
||||
&second,
|
||||
&[("firefox", firefox)],
|
||||
&[
|
||||
("call", call, "peerspeak-owned"),
|
||||
("rec-out", rec_out, "tainted-owner-bridge"),
|
||||
],
|
||||
);
|
||||
}
|
||||
|
||||
// ──────────────────────────────────────────────────────────────────────
|
||||
// Stickiness and lifetime-awareness (v3.4 §6.1.3)
|
||||
// ──────────────────────────────────────────────────────────────────────
|
||||
|
||||
Reference in New Issue
Block a user